Research Journal of Applied Sciences, Engineering and Technology 3(6): 546-552,... ISSN: 2040-7467 © Maxwell Scientific Organization, 2011

advertisement
Research Journal of Applied Sciences, Engineering and Technology 3(6): 546-552, 2011
ISSN: 2040-7467
© Maxwell Scientific Organization, 2011
Received: April 23, 2011
Accepted: May 19, 2011
Published: June 25, 2011
Application of Artificial Neural Network for Modeling Benefit to
Cost Ratio of Broiler Farms in Tropical Regions of Iran
M.D. Heidari, M. Omid and A. Akram
Department of Agricultural Machinery Engineering, Faculty of Agricultural Engineering and
Technology, School of Agriculture and Natural Resources, University of Tehran, Karaj, Iran
Abstract: The economics of poultry meat production depends on numerous factors, but most important is
general economic policy. For the economic analyses, net profit, gross return, net return, Benefit to Cost Ratio
(BCR), productivity, etc. have to be computed. In this study, various Artificial Neural Network (ANN) models
were developed to estimate the BCR of broiler farms in tropical regions of Iran. To develop ANN models, data
were obtained from growers, government officials as well as from relevant databases. The developed ANN was
a Multilayer Feed Forward Network (MLFN) with five neurons in the input layer, one and two hidden layer(s)
of various numbers of neurons and one neuron in the output layer. The MLFN were trained with the
experimental data obtained from 44 broiler farms. Based on performance measures, (5-20-1)-MLFN, namely,
a network having five neurons in its input layer and twenty neurons in the hidden layer resulted in the bestsuited model estimating the BCR. For the optimal model, the values of the model’s outputs correlated well with
actual outputs, with coefficient of determination (R2) of 0.978. For this configuration, MSE, MAE and MAPE
values were 0.002, 0.037 and 2.695, respectively. Sensitivity analysis revealed that feed cost is the most
significant parameter in modeling the BCR to cost ratio in the broiler production.
Key words: Artificial neural networks, benefit to cost ratio, broiler, economic modeling
There has been a vast literature about ANNs,
basically in the empirical field. Since the middle of 1980s,
ANN has been used in economic, energy and
environmental modeling in recent studies. Zhang and Fuh
(1998) utilized ANN to estimate packaging costs based on
product dimensions. This approach has known the first
applications in the manufacturing sector for planning,
emulation and management of production processes and
plants. Ermis et al. (2007) analyzed world green energy
consumption through artificial neural networks. They
analyzed world primary energy including fossil fuels such
as coal, oil and natural gas, using feed forward back
propagation ANN. Zangeneh et al. (2010) calculated
machinery energy ratio of potato production using the
ANN technique under single hidden layer specification.
They used ANN to characterize and assess mechanization
status of potato farms in Hamadan province of Iran with
a view point of energy expenditure in farm machinery.
Azadeh et al. (2008) used multi-layer perceptron of
artificial neural network for forecasting monthly electrical
energy consumption of Iran. Computer simulation is
developed to generate random variables for monthly
electricity consumption. This is achieved to foresee the
effects of probabilistic distribution on monthly electricity
consumption.
INTRODUCTION
Forecasts of agricultural production and prices are
intended to be useful for farmers, governments, and
agribusiness industries. Because of the special position of
food production in a nation's security, governments have
become both principal suppliers and main users of
agricultural forecasts. They need internal forecasts to
execute policies that provide technical and market support
for the agricultural sector (Geoffrey, 1994).
This study presents an application of Artificial Neural
Networks (ANN) to model Benefit to cost ratio of broiler
farms. The ANNs are universal approximations of
functions and have been successfully used in many
research areas: air traffic control, character and voice
recognition, medical diagnosis and research, weather
prediction, etc. But the ANNs have also been extensively
applied to economics and finance (Vellido et al., 1999).
ANNs have their origins in the study of the complex
behavior of the human brain. Historically, McCulloch and
Pitts (1943) introduced simple models with binary
neurons. Then, Rosenblatt (1958) proposed the multilayer structure with a learning mechanism based on the
work of Hebb (1949), the so-called perceptron, and
the first neural networks applications began with
Widrow (1959).
Corresponding Author: Mohammad Davoud Heidari, Department of Agricultural Machinery Engineering, Faculty of Agricultural
Engineering and Technology, University of Tehran, Karaj, Iran. Tel: +98-261-2801038/+98-9131568355, Fax: +98-261-2808138
546
Res. J. Appl. Sci. Eng. Technol., 2(6): 546-552, 2011
When function approximation is the goal, the MLFN
model will often deliver close to the best fit. Omid et al.
(2009) simulated drying kinetics of pistachio nuts using a
MLFN. A comparative study among MLFN and empirical
models was also carried out by authors. They showed that
MLFN model is more accurate than the empirical models.
The present work was motivated in this direction. Here
the same methodology is adopted for selected broiler
farms. The objectives were to model variable costs use for
broiler production farms in Yazd province.
BC =
Productivity =
Case study and data collection: Yazd province with the
area of 7215ha (4.37% of total area of country) located in
the center of Iran within 29º48! to 33º30! latitude and
54º45! to 56º30! longitude. In this study, the data were
collected from 44 broiler farms in six villages from Yazd
province, Iran. Share of this province in broiler farms
within Iran was 5% for 2009 production year, with 577
broiler farms. The production of broiler was about 1988
tons/year in Yazd province (Anonymous, 2009). To
develop ANN model, data were obtained from growers,
government officials as well as from relevant databases.
Data were collected from the farmers by using a face-toface questionnaire method. Farms were randomly chosen
from the villages in the area of study. The sample size was
determined using Neyman method and was calculated as
44 farms (Yamane, 1967).
The economics of poultry meat production depends
on numerous factors, but most important is general
economic policy. Other factors include the choice of
production technology, the organization and the
productivity of labor, and the extent of the exploitation of
the productive factors. Variable costs (direct costs) in
broiler farms included the cost of pullets, feed costs,
water, electricity, health care costs (medicine, disinfection
and vaccinations), labor and etc.
For the economic analyses, net profit, gross return,
net return, benefit to cost ratio and productivity were
computed by (Heidari and Omid, 2011):
Net return =
Total production value ($ (1000
bird)G1) – Total production cost ($
(3)
(1000 bird)G1)
H
I
j=1
i=1
yk = f 2 ( wk 0 + ∑ wkj f1 ( w j 0 + ∑ w ji xi ))
(6)
where xi is the input value to node i of the input layer,
Hj is the hidden value to node j of the hidden layer, and yk
is the output at node k of the output layer (O). An input
layer bias term I0 = 1 with bias weights wj0 and an output
layer bias term H0 = 1 with bias weights wk0 are included
to permit adjustments of the mean level at each stage.
In the MLFNs, error minimization can be obtained by
a number of procedures including Gradient Descent (GD),
Levenberg-Marquardt (LM), and Conjugate Gradient
(CG). MLFN are normally trained with an error backpropagation (BP) algorithm. The knowledge obtained
during the training phase is not stored as equations or in
a knowledge base but is distributed throughout the
network in the form of connection weights between
neurons. It is a general method for iteratively solving for
weights and biases. BP uses a GD technique that is very
stable when a small learning rate is used but has slow
convergence properties. Several methods for speeding up
BP have been used, including adding a momentum term
or using a variable learning rate. In this paper, GD with a
Total production value = Broiler (kg (1000 bird)G1) ×
Price of commodity ($ (1000
(1)
bird)G1)
Total production value ($ (1000
bird)G1) – Variable costs ($ (1000
(2)
bird)G1)
Broiler (kg (1000 bird)G1) / Total
production cost ($ (1000 bird)G1) (5)
Artificial neural network model: An artificial neural
network consists of interconnected identical simple
processing units called neurons. The connections between
neurons are called synapses and could have different
levels of electrical conductivity, which is referred to as
the weight of the connection. This network of neurons and
synapses stores the knowledge in a ‘‘distributed’’ manner:
the information is coded as an electrical impulse in the
neurons and is stored by changing the weight (i.e., the
conductivity) of the connections. Each connection to a
neuron has an adjustable weight factor associated with it.
A Multi-layer Feed-Forward Network (MLFN) consists of
three layers: an input layer, one or more hidden layers and
an output layer. The input from each neuron in the input
layer is multiplied by an adjustable connection weight. At
each neuron, the weighted input signals are summed and
this combined input then passed through a non-linear
transfer function (f) to produce the output of the neuron.
The output of one neuron is the input of the neurons in the
next layer. The process in the MLFN model that consists
of a single hidden layer can be summarized as
(Omid et al., 2009):
MATERIALS AND METHODS
Gross return =
Total production value ($ (1000 bird)G1) / Total
(4)
production cost ($ (1000 bird)G1)
547
Res. J. Appl. Sci. Eng. Technol., 2(6): 546-552, 2011
Table 1: Budgetary analyses of broiler farms production
Cost and return components
Unit
Yield
kg (1000bird)G1
Sale price
$ (1000bird)G1
Gross Value of Production
$ (1000bird)G1
Variable Cost of Production
$ (1000bird)G1
Fixed Cost of Production
$ (1000bird)G1
Total Cost of Production
$ (1000bird)G1
Total Cost of Production
$ kgG1
Gross Return
$ (1000bird)G1
Net Return
$ (1000bird)G1
Benefit/Cost Ratio
Productivity
kg (1000bird)G1
momentum (GDM) algorithm that is an improvement to
the straight GD rule in the sense that a momentum term is
used to avoiding local minima, speeding up learning and
stabilizing convergence, is used (Omid et al., 2009).
Model selection: A program was developed in
NeuroSolutions 5.07 package for the feed forward and
back propagation network. We used the ‘Leave N out’
option, i.e., networks were trained multiple times leaving
out different sections of data for each training run. This
training procedure is very useful for testing the robustness
of a model on small datasets. Here we divided the 44 data
into eleven (N = 11) subsets of equal size. We then
trained the net N times, each time leaving out one of the
subsets from training, but using only the omitted subset to
compute error criterion.
To objectively evaluate the performance of the
network, four different statistical indicators were used.
These indicators are - mean squared error (MSE), mean
absolute error (MAE), coefficient of determination (R2)
and mean absolute percentage error (MAPE) (Zangeneh
et al., 2010):
(
1 n
−Y
∑ Y
n i = 1 Estimated T arg et
MSE =
n
R2 =
(
∑ YEstimated − YT arg et
i=1
n
)
∑ ( YEstimated − YMean )
)
2
it is important to have a good idea of how much feed is
eaten, in particular the amount of feed needed per kg of
meat. That is called the feed conversion. The total
expenditure for the broiler production was 3648.47 $
(1000bird)G1 and with attention to the gross production
value (5035 $ (1000bird)G1), gross return were found to be
1528.71 $ (1000bird)G1. With respect to results of Table
1, the Benefit-Cost Ratio (BCR) from broiler production
in the surveyed farms was calculated to be 1.38. Similar
results for BCR can be seen in study of Heidari and Omid
(2011) for greenhouse cucumber as 1.68. Depending on
the farm size, broiler farming can be a main source of
family income or can provide subsidiary income and
gainful employment to farmers throughout the year.
For this study, ANN represents a valid tool for the
identification of the transfer function of the analyzed
processes, through an implicit link between the input
values (variable costs of broiler production) and the
output value (BCR).
Various MLFN were designed and trained as two and
three layers to find an optimal model prediction for the
Benefit to cost ratio. Training procedures of the networks
was as follows: Different hidden layer neurons and
arrangements were adapted to select the best production
results (Table 2). Altogether, 20 configurations with
different number of hidden layers (varied between one
and two), different number of neurons for each of the
hidden layers, and different inter-unit connection
mechanisms were designed and tested. Specifically, the
number of neurons in one hidden layer network was
varied from 2 to 20, whereas in two hidden layers cases
were varied from 2 to 8 (in the first) and 2 to 12 (in the
second) hidden layer, respectively. For illustration
purposes, in Fig. 1 a network having one hidden layer
and one output layer is drawn. We name this network a
(5-3-1)-MLFN topology; that is, I = 5, j = 3 and k = 1.
Therefore, ANNs with five inputs and one output
have been trained to estimate network parameters. The
simulated data (variable costs) are used to train the
feedforward neural networks using GDM algorithm.
Results of the MLFN trained for different arrangement are
presented in Table 2. As indicated in Table 2, among the
trained networks, the (5-20-1)-MLFN, namely, a network
(7)
2
2
(8)
i=1
MAPE =
1 N YEstimated − YT arg et
100)
∑(
N i=1
YT arg et
Broiler
2601.82
1.92
5035.00
3506.29
142.18
3648.47
1.40
1528.71
1386.53
1.38
0.73
(9)
where YTarget and YEstimated are the actual and estimated
benefit to cost ratio values of ith farm, respectively, YMean
is the mean of actual Benefit to cost ratio values, and N is
the number of observations.
RESULTS
Average capacity of surveyed farms was 18142 birds.
Maximum, minimum and average meat production of
farms was 3000, 2000 and 2601 kg (1000bird)-1,
respectively. The total cost of broiler production and the
gross value of its production were calculated and shown
in Table 1.
The fixed and variable expenditures included in the
cost of production were calculated separately as 142.18
and 3506.29 $ (1000bird)G1, respectively. Fixed costs
included only 3.9% of total costs. Feed costs are, as a
rule, the greatest expenditure of a broiler farm. Therefore,
548
Res. J. Appl. Sci. Eng. Technol., 2(6): 546-552, 2011
Table 2: Alternative configurations of ANN for Benefit to Cost Ratio (BCR) of broiler farms
ANN Model
NH1
NH2
MSE
MAE
1
2
0
0.006
0.063
2
3
0
0.005
0.058
3
5
0
0.005
0.054
4
7
0
0.005
0.062
5
8
0
0.005
0.058
6
9
0
0.002
0.038
7
11
0
0.003
0.050
8
15
0
0.002
0.040
9
17
0
0.002
0.030
10
20
0
0.002
0.037
11
2
2
0.006
0.067
12
2
4
0.006
0.066
13
3
4
0.006
0.059
14
3
5
0.004
0.054
15
4
3
0.007
0.071
16
4
4
0.007
0.067
17
5
4
0.006
0.062
18
5
8
0.005
0.055
19
6
8
0.003
0.052
20
8
12
0.003
0.050
*: Optimum configuration
MAPE
4.577
4.256
3.880
4.488
4.261
2.770
3.586
2.906
2.698
2.695
4.838
4.763
4.260
3.880
5.162
4.885
4.531
4.011
3.724
3.565
R2
0.630
0.514
0.616
0.542
0.507
0.585
0.875
0.783
0.879
0.978
0.895
0.515
0.448
0.442
0.665
0.441
0.423
0.590
0.562
0.744
Fig. 1: The (5-3-1)-MLFN topology
having five input variables, twenty neurons in the hidden
layer and single output variable resulted in the best-suited
model estimating the benefit to cost ratio. The
performance of various MLFN models is presented in
Table 2. The coefficient of determination (R2) between the
output of the ANN model (estimated) and the actual
(calculated) value of benefit to cost ratio indicator was
0.978 as highlighted in Table 2 and Fig. 2. For this
configuration, MSE, MAE and MAPE values were 0.002,
0.037 and 2.695, respectively.
Figure 3 shows MAPE of the output indicator for
validation pattern, obtained by comparing the outputs of
the best ANN model of desired outputs and the actual
outputs. The validation subset contained ten patterns that
were not used in the training and testing phases. Figure 2
shows the correlation between the model’s outputs and
calculated output. In general, one hidden layer networks
having eleven to twenty hidden units showed better
performance. Among these, a single hidden layer with
twenty neurons was finally selected, because the number
549
Res. J. Appl. Sci. Eng. Technol., 2(6): 546-552, 2011
1.00
0.90
R2
0.80
0.70
0.60
0.50
0.40
1
3
2
4
5
6
7
8
9 10 11 12 13
Number of ANN model
14
15
16
17
18
19
20
Fig. 2: Correlation between the ANN model’s outputs and calculated outputs
5
4
3
MAPE (%)
2
1
0
-1
-2
-3
-4
-5
1
2
3
4
8
5
6
7
Testing data set (Nember of farms)
9
10
11
Fig. 3: MAPE in the benefit to cost ratio (BCR) estimation over the test set data
0.050
Sensitivity
0.040
0.030
0.020
0.010
0.000
Chick cost
Labor cost
Feed cost
Fuel cost
Electricity cost
ANN inputs
Fig. 4: Sensitivity analysis of ANN inputs on benefit to cost ratio (BCR)
of hidden units should
(Zhang et al., 1998).
be
as
electricity (for BCR prediction). Diesel fuel had the small
share of variable costs while it had the highest energy
consumption. This is because of low price of fuel and
subsidizing policy in Iran. The average of 2314.49 l of
diesel fuel was consumed for the heating production
rooms.
Electricity power was used in automatic feeding and
lighting equipment's. Artificial lighting is the way to raise
few as possible
DISCUSSION
Analysis of selected parameters: The parameters that we
used for the MLFN models were included: costs of
different sources such as chick, labor, feed, fuel and
550
Res. J. Appl. Sci. Eng. Technol., 2(6): 546-552, 2011
Table 3: Sensitivity analysis results for the variables considered in ANN analysis
Variable No
Variable name
Rank
1
Chick cost
4
2
Labor cost
3
3
Feed cost
1
4
Fuel cost
5
5
Electricity cost
2
the production of chickens. If the housing is lit in the
cooler hours before sunrise or after sunset, the chickens
are able to eat more.
C
Sensitivity analysis of variable costs: Parallel with
MLFN statistics, also ANN sensitivity analysis for input
variables was performed (Table 3, Fig. 4). In order to
assess the predictive ability and validity of the developed
MLFN model, a sensitivity analysis was performed using
the best network selected. The robustness and sensitivity
of the model were determined by examining and
comparing the output produced during the validation stage
with the calculated values. The MLFN model was trained
by withdrawing each input item one at a time while not
changing any of the other items for every pattern.
According to obtained results in Fig. 4, the share of
each input item of developed MLFN model on desired
output can be seen clearly. Sensitivity analysis provides
insight into the usefulness of individual variables. With
this kind of analysis it is possible to judge what
parameters are the most significant (with sensitivity value
close to 1) and the least significant (with sensitivity value
close to 0) during generation of the satisfactory MLFN.
According to sensitivity analysis the degree of feed cost
according to value of 0.043 is the most significant
parameter for MLFN models in modeling the benefit to
cost ratio in the broiler production. Less significant appear
to be electricity, labor, chick and fuel costs as: 0.043,
0.024, 0.013 and 0.007, respectively.
C
Sensitivity analysis
0.013
0.024
0.043
0.007
0.043
Sensitivity analysis revealed that among the most
significant predictive variables, feed cost and
electricity cost were identified, which are often
recognized as sensitive factors.
Estimation of average cost of production and pricing
based on this information can reduce tensions in
poultry market. Government with this information
can forecast the future price of poultry meat, adjust
its market situation and buy potatoes in excess to
market requirements.
Nomenclature and abbreviations:
ANN
Artificial neural network
BCR
Benefit to cost ratio
BP
Back-propagation
CG
Conjugate gradient
GD
Gradient descent
LM
Levenberg-Marquardt
MLP
Multi layer perceptron
MAPE Mean absolute percentage error
MLFN Multilayer feedforward network
MSE
Mean squared error
MAE Mean absolute error
ACKNOWLEDGMENT
The authors appreciate the support from the AgriJihad Organization of Yazd province, Iran. The financial
support provided by University of Tehran, Iran, is duly
acknowledged.
CONCLUSION
REFERENCES
This paper shows the valuable application of Multilayer Feed Forward Network (MLFN) in modeling the
variable costs use in the broiler production in the central
region of Iran. Summary of conclusion can be stated as
follows:
C
C
Anonymous. 2009. Annual Agricultural Statistics.
Ministry of Jihad-e-Agriculture of Iran.
http://www.maj.ir.
Azadeh, A., S.F. Ghaderi and S. Sohrabkhani, 2008. A
simulated-based neural network algorithm for
forecasting electrical energy consumption in Iran.
Energ. Policy, 36: 2637-2644.
Ermis, K., A. Midilli, I. Dincer and M.A. Rosen, 2007.
Artificial neural network analysis of world green
energy use. Energ. Policy, 35: 1731-1743.
Geoffrey, A.P., 1994. Economic forecasting in
agriculture. Int. J. Forecasting, 10(1): 81-135.
Hebb, D.O., 1949. The Organization of Behavior. Science
Edn., New York.
Heidari, M.D. and M. Omid, 2011. Energy use patterns
and econometric models of major greenhouse
vegetable productions in Iran. Energy, 36: 220-225.
Broiler production consumed a total cost of 3648.47
$ (1000 birdG1), which was mainly due to feed.
A 5-20-1 MLFN namely, a network having five
neurons in the input layer and twenty neurons in the
hidden layer, resulted in the best-suited model
estimating the benefit to cost ratio. This MLFN
architecture was capable of ‘‘learning” the
relationship between the input pretreatment process
parameters (variable costs) and output pretreatment
process parameter (benefit to cost ratio) for studied
broiler farms as indicated by low MAPE (2.695%)
and MSE (0.002).
551
Res. J. Appl. Sci. Eng. Technol., 2(6): 546-552, 2011
McCulloch, W.S. and W. Pitts, 1943. A logical calculus
of the ideas immanent in nervous activity. Bull. Math
Biophys., 5: 115-133.
Omid, M., A. Baharlooei and H. Ahmadi, 2009. Modeling
drying kinetics of pistachio nuts with multilayer feedforward neura l network. Dry. Technol., 10:
1069-1077.
Rosenblatt, F., 1958. The perceptron: a probabilistic
model for information storage and organization in the
brain. Psychol. Rev., 62: 386-408.
Vellido, A., P.J.G. Lisboa and J. Vaughan, 1999. Neural
networks in business: A survey of applications
(1992-1998). Expert Syst. Appl., 17: 51-70.
Widrow, B., 1959. Adaptative sampled-data systems, a
statistical theory of adaptation, 1959 IRE WESCON
Convention Record. Part 4, Institute of Radio
Engineers, New York.
Yamane, T., 1967. Elementary Sampling Theory. Prentice
Hall Englewood Cliffs, New Jersey, USA.
Zangeneh, M., M. Omid and A. Akram, 2010.
Assessment of machinery energy ratio in potato
production by means of artificial neural network. Afr.
J. Agric. Res., 5: 993-998.
Zhang, G., B.E. Patuwo and M.Y. Hu, 1998. Forecasting
with artificial neural networks: The state of the art.
Int. J. Forecasting, 14: 35-62.
Zhang, Y.F. and J.Y. Fuh, 1998. A neural network
approach for early cost estimation of packaging
products. Comput. Ind. Eng., 34(2): 433-450.
552
Download